Watermark embedding: Hiding a signal within a cover image

When looked at as a communication task, the watermarking process can be split into three main steps: watermark generation and embedding (information transmission), possible attacks (transmission through the channel), and watermark retrieval (information decoding at the receiver side). In this article we review the main issues in watermark generation and embedding. By focusing on the case of image watermarking, we first discuss the choice of the image features the watermark is superimposed to. Then we consider watermark generation and the rule used to insert the watermark within the host features. By adopting again a communication perspective, some useful hints are given on the way the watermark should be shaped and inserted within the host document for increased robustness against attacks. Given that invisibility is one of the main requirements a watermark must satisfy, the way psycho-visual notions can be used to effectively hide the watermark within an image is carefully reviewed in the second part of the article. Rather than insisting on the mathematical aspects of each of the above issues, the main rationale behind the most commonly adopted approaches is given, as well as some illustrative examples.